Every agent knows its part.
Spec-driven coordination for AI-assisted development.
Define work as specs. Run it anywhere — locally, or on Synchestra Hub.
Spec-driven coordination for AI-assisted development.
Define work as specs. Run it anywhere — locally, or on Synchestra Hub.
Running one agent on one task works. Running five agents across three platforms on a tree of interdependent tasks? That's where things fall apart.
Every time an agent starts, it needs to understand what's going on. Loading the full project context burns tokens and loses focus.
What's been done? What's left? What's blocked? The answers live in chat logs, git history, and your head.
Agent A finishes a sub-task. Agent B needs the output. Without a shared protocol, you're the glue.
Agents work across platforms asynchronously. There's no dashboard, no progress bar, no single place to look.
When five agents work across three platforms, the answers to "what's done?" and "what's next?" need to live somewhere every tool can reach. Synchestra turns a git repository into a coordination protocol. The repo is the database. Git history is the audit trail.
Features have sub-features. Tasks have sub-tasks. The directory tree mirrors how work actually decomposes.
Claude Code, Cursor, GPT, custom scripts — any agent that can read files can participate.
Auto-generates minimal per-task context. Agents stay focused, costs stay low.
Agents claim tasks atomically through git. Optimistic locking, schema-validated at every commit.
Install the CLI to work locally, or connect via the web to manage projects across machines. A specscore-spec-repo.yaml links everything together.
Create tasks from specs, plans, or freeform descriptions. Agents discover and claim them asynchronously using optimistic locking through git.
Every task has a status. Every agent has a log. Watch work unfold in real time across platforms and agents.
Synchestra doesn't replace your agent runtime. It's the layer above — defining what needs doing so agents can focus on how.
The specification format is SpecScore — an open-source framework for writing specs that AI agents can understand and execute. Features, plans, architecture, project structure — SpecScore defines the format. Synchestra adds orchestration, tasks, and coordination on top.
Use SpecScore standalone with any tooling you already have — or pair it with Synchestra for the full spec-to-execution workflow.
| Traditional orchestrators | Synchestra | |
|---|---|---|
| State storage | Database / API server | Git repository |
| Agent integration | SDK / API client | CLI, MCP, or direct file access |
| Infrastructure | Server, database, networking | Git + single binary |
| Context loading | Full project dump | Auto-generated minimal context per task |
| Multi-platform | Locked to one runtime | Any tool that can read/write files |
| Coordination | Real-time messaging | Async via repo state |
| Audit trail | Event log in database | Git history |
| Validation | Application-level checks | Schema-enforced at commit time |
Agent frameworks are runtimes — they execute agents and chain tool calls. Synchestra is the coordination layer above. Your agents across any platform coordinate through the same repo.
Linear, Jira — those track work for humans. Synchestra tracks work for agents and humans. Machine-navigable and human-readable.
CI/CD pipelines are linear and event-driven. Synchestra workflows are hierarchical and async. Tasks spawn sub-tasks dynamically.
You could do what Synchestra does manually — like you could deploy without CI/CD or track bugs without an issue tracker. Synchestra is the automation and discipline layer that makes it sustainable. Schema validation, context generation, conflict resolution, progress tracking — the chain of small steps nobody has time to do consistently.
They don't break silently. The CLI validates on every operation, pre-commit hooks catch problems before they enter the repo, pre-push hooks catch anything that slipped through, and GitHub Actions provide a final safety net.
Through optimistic locking — the same mechanism distributed systems have used for decades. Agents claim tasks by committing and pushing. If the push fails, someone else got there first. For the remaining edge cases, Synchestra provides AI-powered merge resolution. The "commit often" philosophy minimizes the conflict window in the first place.
Git already handles distributed collaboration. Schema validation works regardless of team size. Optimistic locking handles concurrency without a central coordinator. The same conventions that keep a solo developer organized keep a team aligned.
The best test of coordination infrastructure is using it yourself. Synchestra's own development — across three repos, multiple agents, and daily commits — is managed by Synchestra. The codebase is small by design — every rough edge felt immediately, fixed quickly, and built in the open.
One command. Detects your OS and architecture, downloads the latest release, verifies the checksum, and installs the binary.
curl -fsSL https://synchestra.io/get-cli | sh
Linux and macOS, amd64 and arm64. Windows archives available from releases.
SHA-256 checksums are downloaded and verified automatically before installation.
Single static binary. No runtime, no package manager, no containers.
You run everything locally with your agent of choice. And you probably should — it's fast, it's private, it works.
But then you need to close your laptop. The agent isn't done. You wish it could keep going without you, but it can't. The work stops when you stop.
Synchestra Hub picks up where your laptop leaves off.
Run agents locally, on a remote VM, or in the cloud. Transfer work and context mid-flight. Come back to finished tasks, not stale sessions.
Dispatch agents to VMs or cloud infrastructure. Self-hosted or managed — your choice.
See what every agent is doing across every project. Task progress, logs, and status — all in one place.
Queue tasks, trigger workflows, and connect to your existing tools through the Hub API.
Synchestra Hub builds on the same specs and CLI you already use. No lock-in — the open-source foundation works independently.
Get started in minutes. No infrastructure required.
Explore the source code & specifications →